An application of general regression neural network to nonlinear adaptive control

Neural networks have the potential to learn multivariable static mappings via the adjustment of internal weights. Therefore they are able to form a self-organizing control structure in order to handle unknown or slowly varying plant parameters and nonlinearities. In this paper it is shown that the general regression neural network can be applied to a broad class of such systems. The feasibility of the approach is demonstrated with a second order plant with unknown nonlinearity and unknown PT₁ parameters in order to perform input-output linearization. The neural network interacts with the plant to estimate and compensate the nonlinearity and PT₁ parameters. The learning scheme is fed by error signals generated by a comparison between the states of a reference model and the actual plant. It can be demonstrated that the overall system is stable in the sense of Luapunov. The attractive features of this approach are the high speed of the GRNN implemented in parallel hardware and the ability for constant learning in a changing environment